04 October 2001
Intelligent sensor networks for plant floor or remote, hostile environments
by Bob Felton
Forget about configuration problemsthe new sensors will do it themselves.
Plant engineers, foresters, military commandersanybody who needs to collect physical data distributed across large areaswill soon be able to create ad hoc sensor networks that decide for themselves how to configure individual sensors. "In extreme cases," according to a paper presented at the Sixth International Symposium on Communications Theory and Applications, "nodes may be dropped from an aircraft in a remote terrain; however, even assuming individual placement, the scale of the system and variations in the environment require that they self-configure and adapt to their environment without user intervention."
Dropped from an airplane over the mountains of western North Carolina, they might have helped the FBI locate alleged terrorist Eric Rudolph. Strewn across a battlefield, they’ll help protect troops against biological warfare agents. At accident scenes, rescue workers will use them to scout and help negotiate their way through rubble. On the plant floor, they’ll help maintain equipment, generate real-time maps of temperature or other environmental variables, and provide physical security.
As with so much other technology, it is the Defense Advanced Research Projects Agency (DARPA) and the National Science Foundation (NSF) that are putting up most of the money to develop self-configuring sensor networks. When fully realized, military commanders will be able to blanket an area with sensors dropped from aircraft, which will then organize themselves into a network and get busy. If something destroys or moves some of the sensors, the remaining sensors will simply reconfigure themselves and continue working.
Invisible hand architecture
Traditionally, whether communications are wireless or tethered, distributed sensor networks operate under a centralized, top-down approach. The new approach aims to create a software-based invisible hand similar to Adam Smith’s. "We have hypothesized," the researchers said, "that sensor network coordination applications are better realized using localized algorithms. We use this term to mean a distributed computation in which sensor nodes only communicate with sensors within some neighborhood, yet the overall computation achieves a desired global objective."
The advantages of distributed computation, the researchers said, are significant. "Because each node communicates only with other nodes in some neighborhood, the communication overhead scales well with increase in network size. For a similar reason these algorithms are robust to network partitions and node failures." Neighborhood, in this case, means sensors within the range of a tiny, low-power, onboard radio transmitter.
Almost the first question a sensor must answer in real-time improvisation of a network is: "Where am I?" Further, to conserve power and because global positioning system data may not always be available, the designers are developing a system that relies on building "a coordinate system based on sensory measurements of the physical environment" such as signal strength, signal propagation characteristics, or packet delivery rates. In one approach, the researchers explained, "Sensor nodes measure a sufficient number of pairwise distance estimates and then use multilateration algorithms for position estimation."
Not every installation has the same requirements, though. "It is difficult," the researchers acknowledged, "to design localized algorithms that both empirically adapt to a wide range of environments and converge to a desired global behavior over that entire range. Some information about the system can significantly help the convergence of localized algorithms. External system information may be provided in several ways."
Simulations have demonstrated, for instance, that identifying boundary nodes enhances system life because you can program adjoining, interior nodes to use less broadcast power. IT